Scholarlink

<p>Currently, object detection models utilized in UAV aerial image tasks encounter challenges such as small and dense objects, as well as interference from complex backgrounds. This paper introduces ADD-YOLO, an improved model based on YOLOv8s. In this model, the traditional convolutional layer is replaced with AKConv, which enhances the model’s ability to adapt to object variations. The C2f_DRAC structure, integrated with AKConv and CBAM, improves the model’s ability to capture multi-scale contextual information and effectively handles background interference. The DABFPN structure includes a dedicated small object detection layer, which improves performance in detecting small objects and addresses challenges posed by background interference. Furthermore, CIoU-Soft-NMS replaces the original NMS, improving the detection of dense objects and resolving issues such as the incorrect elimination of adjacent prediction boxes and inaccuracies in IoU calculations due to overlapping bounding boxes. Extensive ablation studies and comparative experiments were conducted on the VisDrone2019 dataset and the UAVDT benchmark. The results demonstrate that ADD-YOLO outperforms the leading models in UAV aerial image detection tasks, achieving improvements of 15.7% and 7.3% in mAP@0.5 and 13.8% and 5.1% in mAP@0.5:0.95, respectively, thereby validating the effectiveness of this model.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

ADD-YOLO: a new model for object detection in aerial images

  • 期刊论文

  • 发表日期：

    2025年2月18日

查看全文
  • Yifei Yang,
  • Zhengyong Feng,
  • Wei Jin,
  • Pengcheng Miao

摘要

Currently, object detection models utilized in UAV aerial image tasks encounter challenges such as small and dense objects, as well as interference from complex backgrounds. This paper introduces ADD-YOLO, an improved model based on YOLOv8s. In this model, the traditional convolutional layer is replaced with AKConv, which enhances the model’s ability to adapt to object variations. The C2f_DRAC structure, integrated with AKConv and CBAM, improves the model’s ability to capture multi-scale contextual information and effectively handles background interference. The DABFPN structure includes a dedicated small object detection layer, which improves performance in detecting small objects and addresses challenges posed by background interference. Furthermore, CIoU-Soft-NMS replaces the original NMS, improving the detection of dense objects and resolving issues such as the incorrect elimination of adjacent prediction boxes and inaccuracies in IoU calculations due to overlapping bounding boxes. Extensive ablation studies and comparative experiments were conducted on the VisDrone2019 dataset and the UAVDT benchmark. The results demonstrate that ADD-YOLO outperforms the leading models in UAV aerial image detection tasks, achieving improvements of 15.7% and 7.3% in mAP@0.5 and 13.8% and 5.1% in mAP@0.5:0.95, respectively, thereby validating the effectiveness of this model.